At the first World Robot Taishan Competition held recently in Tai'an, Shandong Province, the Yushu G1 development version of the robot, the same model used in the Spring Festival Gala, encountered problems due to machine overheating. For safety reasons, the operator chose to withdraw from the competition.
"Yushu Robot withdrew from the Taishan Climbing Competition due to overheating" became a hot search topic on Weibo, attracting public attention to the actual working capabilities of humanoid robots. Why does the flexible robot on the stage suffer from heat stroke when going up the mountain? How far is it from practical application? A reporter from Science and Technology Daily interviewed relevant experts in the industry.
The first question: Why does a robot suffer "heatstroke" when climbing?
An expert from the China Electronics and Information Industry Development Research Institute who did not want to be named said that this competition is more like a test, testing whether humanoid robots can convert short-term movement capabilities into continuous operation capabilities.
Experts explained that the so-called "heatstroke" of robots is mainly because the heat generated by the body cannot be discharged in time, not because the outside temperature is very high. Stage performances are usually performed in a relatively flat and controllable environment. The movements are repeatedly trained and choreographed, with the focus being on performing a difficult movement. However, the stone steps of Taishan require the robot to judge the foothold while constantly lifting its own weight. Hips, knees, ankles and other joints must output greater force for a long time. Especially when the load is high, the speed is low, and the posture is frequently adjusted, the joint motor is more likely to accumulate heat.
The heat dissipation method in the public parameters of Yushu G1 is local air cooling, and the nominal battery life is about two hours, but the nominal battery life cannot be directly equated to the working time of continuous mountain climbing. The robot has indeed been able to demonstrate a high upper limit of movement, but maintaining its ability stably for tens of minutes or even longer under real load is still a difficulty that needs to be overcome.
At a deeper level, it's an engineering trade-off between flexibility and the ability to work continuously. In order to make the robot move briskly, developers need to pack motors, reducers, and sensors into compact joints as much as possible; but to allow it to climb hills for a long time, it requires stronger continuous output capabilities, better heat dissipation, and more power, which often increases volume, weight, and energy consumption.
Second question: What is more important when measuring the value of a robot?
Zhang Weimin, a professor at Beijing Institute of Technology and chairman of Beijing Institute of Technology Huahui Intelligent Technology Co., Ltd., believes that the most obvious shortcoming exposed in this competition is that the robot's ability to autonomously adapt when facing irregular terrain is still insufficient. The height, depth, slope and friction conditions of Taishan's steps are constantly changing. Every time the robot takes a step, it needs to judge whether its feet are stable, and then adjust its center of gravity according to the actual force after landing. The stage actions can be pre-arranged, but the real mountain roads will not appear according to the conditions during training. Some robots staggered, retreated, and paused frequently, reflecting that the cooperation between environmental perception, landing planning, contact force control, and whole-body balance was not mature enough. The subtle adjustments that humans make at any time when walking on mountain trails are still very complicated for robots.
Zhang Weimin believes that what deserves more attention is the ability of independent error correction and safe handling. There is a robot on site that needs to be pulled by an operator to move forward, indicating that it still relies on people's immediate help on difficult sections. Robots that are truly put into production and life must not only make the right moves when everything is normal, but also make independent judgments about slowing down, changing gait, and re-planning routes when they slip, miss the mark, have their paths blocked, or have joint temperatures rise. Having someone to support you at all times to reach the finish line and completing the task independently and returning safely represent two different technical levels.
"Of course we can't just see the 'rollover' scene, there are also robots that have performed outstandingly in competitions." The unnamed expert emphasized that in the future, measuring the practical value of a robot will not just look at how fast it runs or how high it jumps, but also how long it continues to work, the number of tasks, the number of manual interventions, and whether it can recover on its own after an accident. What the industry really needs to strive for is stable and reliable effective working hours.
Question 3: How far is it from practical application?
"The biggest bottleneck of humanoid robots is not the 'brain', but the ability to continue to operate autonomously." Wang Peng, an associate researcher at the Beijing Academy of Social Sciences, said that the robot's "brain" can already understand the environment, understand instructions, and learn complex actions, but motion control, hardware and task decision-making have not yet been truly coordinated. When climbing a mountain, the joints heat up. It must not only determine where to step next, but also whether to slow down, change gait, or even rest. This requires the coordinated efforts of intelligent models, motion control, actuators, thermal management, and battery management.
Currently, embodied intelligence is shifting from completing specified actions to adapting to unfamiliar environments. "However, there is still a gap between completing limited tasks and handling exceptions autonomously for a long time." Wang Peng said, especially as the home environment continues to change, and the elderly, children, and pets may suddenly enter the work area, which requires higher security and fault tolerance.
Therefore, Wang Peng said that it is more feasible to promote humanoid robots in different scenarios and stages. The factory has clear boundaries for fixed workstations, storage and transportation, and closed park inspections. Charging, maintenance, and manual takeover are easy to arrange. Stable positions are best established first. In terms of emergency rescue, quadruped and crawler robots are more adaptable to some terrains, and it is not suitable to blindly pursue humanoid robots. What is more realistic in the near future is to combine manual collaboration, remote operation and local autonomy.
"In the next two to three years, embodied intelligence is more likely to become a reliable helper in a number of specific positions. It will take longer for general-purpose humanoid robots to independently handle housework and undertake high-risk rescue operations." The unnamed expert mentioned above said that the key to industrial competition will also shift from "how many actions it can perform" to "how many valuable tasks it can reliably complete per hour."
Reporter Fu Lili
